Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!rutgers!ucla-cs!zen!ucbvax!IBM.COM!ELINSKY
From: ELINSKY@IBM.COM (Jay Elinsky)
Newsgroups: comp.protocols.tcp-ip
Subject: Source routing
Message-ID: <073087.233339.elinsky@ibm.com>
Date: Fri, 31-Jul-87 17:43:21 EDT
Article-I.D.: ibm.073087.233339.elinsky
Posted: Fri Jul 31 17:43:21 1987
Date-Received: Sun, 2-Aug-87 03:14:46 EDT
Sender: daemon@ucbvax.BERKELEY.EDU
Distribution: world
Organization: The ARPA Internet
Lines: 38

Here's a copy of a note I sent to IBMTCP-L, on the subject of source routing
in the IBM TCP/IP code.  My apologies in advance if it shows up twice --
someone else may have also sent it, but it hasn't shown up on our database at
Yorktown.  I want to make sure it shows up on this mailing list, since the
note that started it all appeared here.

Date:     Wednesday, 29 Jul 1987 17:54:27 EDT
From:     Jay Elinsky <ELINSKY@YKTVMX>
To:       <IBMTCP-L@CUNYVM>
Subject:  Source routing

I'm one of the developers of the VM code at Yorktown, and I want to
clarify some possible misconceptions on this subject:

1) We use source routing only on Token Ring.  Even on Token Ring, we can
certainly also route through a router, if the destination host is on
a different net or subnet.  If the destination host is on a different
ring with the same net or subnet number, we do indeed use source routing.
I believe that this is consistent with the architecture of the Token
Ring.  Is it inconsistent with any existing implementation of TCP/IP
over Token Ring?  If so, I'd certainly like to know about it.

2) Re "No RIF field, no function":  If a host is on the same ring, the
ARP code will recognize it, and subsequent packets between the hosts have
no routing field at all.

3) Our TCP/IP certainly is inter-operable with other TCP/IP implement-
ations.  That is our reason for supporting TCP/IP. To repeat what I
said in (1), the source routing issue is only on Token Ring, and to
the best of my knowledge our implementation is consistent with other
TCP/IP Token Ring implementations, and with Token Ring architecture.

4) In the VM code, the Token Ring routing information is kept as part
of the address translation cache.  I think that's logical.

                                       Jay Elinsky
                                       IBM T.J. Watson Research Center
                                       Yorktown Heights, NY